Puzzled Jay

silly stuff, info, personal takes. | art acct: https://puzzledjaypros.tumblr.com | follow @restfulhaven for peaceful scrolling

[doctorvalentlne]

BASED ON THIS POST HERE!->

[acrylicbrushstroke]

if you’re rude to waiters at restaurants i hate you

[catilinas]

i love shipping magazines and i especially love them when they sound like they were written by a mildly aggravated cargo ship

[salix03]

this kingdom is going to shit man

[fjordojustice]

An honorable mention that lives forever in my heart

[ministarfruit]

I love this text post so I drew it

[glowcowboy]

i hope everything gets easier soon. or at least funnier. amen

[adventures-in-poor-planning]

[jmtorres]

do y'all remember, a year into covid, when the evergreen ship got wedged sideways in the suez canal and it was like. oh no blocked shipping route! companies losing gazillions of money! but at least it was a news story whose absurdity causes no deaths? so we were all fixated for a week?

[gotouda]

holy shit. this guy on the bus talking to a girl about how he trades stocks. and she goes “have you seen american psycho” and he said no i dont watch movies. im too busy trading

[satelliteduster]

[zmpl]

has anyone noticed that after the porn ban of 2018 tumblr was essentially killed from the mainstream and everyone flocked to other social media sites like twitter and meta. then those sites got enshittified to where twitter became Nazi Central and meta sites had an entire meme around getting “zucced” aka mark zuckerberg himself would ban you for saying a no-no word like fuck. and then the mainstream shifted to tiktok where infamous toddlerspeak sentences like “he got unalived by a pew pew” were born because if you once again say a no-no word like kill or gun or any other word that isn’t corporate i mean kid friendly then the algorithm will bury your post into the ground. and somehow we’ve come full circle and tumblr is now the most bearable social media site because although we can’t have female presenting nipples we can at least talk to each other like adults. has anyone noticed that at all or is it just me and the flaming skull

[girlvasectomy]

[dappermouth]

do you think sperm whales get scared down there. in the so dark deeps.

[klfg]

[source]

[klfg]

🚨BREAKING: OpenAI published a paper proving that ChatGPT will always make things up.

Not sometimes. Not until the next update. Always. They proved it with math.

Even with perfect training data and unlimited computing power, AI models will still confidently tell you things that are completely false. This isn't a bug they're working on. It's baked into how these systems work at a fundamental level.

And their own numbers are brutal. OpenAI's o1 reasoning model hallucinates 16% of the time. Their newer o3 model? 33%. Their newest o4-mini? 48%. Nearly half of what their most recent model tells you could be fabricated. The "smarter" models are actually getting worse at telling the truth.

Here's why it can't be fixed. Language models work by predicting the next word based on probability. When they hit something uncertain, they don't pause. They don't flag it. They guess. And they guess with complete confidence, because that's exactly what they were trained to do.

The researchers looked at the 10 biggest AI benchmarks used to measure how good these models are. 9 out of 10 give the same score for saying "I don't know" as for giving a completely wrong answer: zero points. The entire testing system literally punishes honesty and rewards guessing.

So the AI learned the optimal strategy: always guess. Never admit uncertainty. Sound confident even when you're making it up.

OpenAI's proposed fix? Have ChatGPT say "I don't know" when it's unsure. Their own math shows this would mean roughly 30% of your questions get no answer. Imagine asking ChatGPT something three times out of ten and getting "I'm not confident enough to respond." Users would leave overnight. So the fix exists, but it would kill the product.

This isn't just OpenAI's problem. DeepMind and Tsinghua University independently reached the same conclusion. Three of the world's top AI labs, working separately, all agree: this is permanent.

Every time ChatGPT gives you an answer, ask yourself: is this real, or is it just a confident guess?

[meinenaffenhosen]

For those that like to actually read the paper : https://arxiv.org/pdf/2509.04664

Why Language Models Hallucinate

Spoiler:

[klfg]

[source]

[klfg]

🚨BREAKING: OpenAI published a paper proving that ChatGPT will always make things up.

Not sometimes. Not until the next update. Always. They proved it with math.

Even with perfect training data and unlimited computing power, AI models will still confidently tell you things that are completely false. This isn't a bug they're working on. It's baked into how these systems work at a fundamental level.

And their own numbers are brutal. OpenAI's o1 reasoning model hallucinates 16% of the time. Their newer o3 model? 33%. Their newest o4-mini? 48%. Nearly half of what their most recent model tells you could be fabricated. The "smarter" models are actually getting worse at telling the truth.

Here's why it can't be fixed. Language models work by predicting the next word based on probability. When they hit something uncertain, they don't pause. They don't flag it. They guess. And they guess with complete confidence, because that's exactly what they were trained to do.

The researchers looked at the 10 biggest AI benchmarks used to measure how good these models are. 9 out of 10 give the same score for saying "I don't know" as for giving a completely wrong answer: zero points. The entire testing system literally punishes honesty and rewards guessing.

So the AI learned the optimal strategy: always guess. Never admit uncertainty. Sound confident even when you're making it up.

OpenAI's proposed fix? Have ChatGPT say "I don't know" when it's unsure. Their own math shows this would mean roughly 30% of your questions get no answer. Imagine asking ChatGPT something three times out of ten and getting "I'm not confident enough to respond." Users would leave overnight. So the fix exists, but it would kill the product.

This isn't just OpenAI's problem. DeepMind and Tsinghua University independently reached the same conclusion. Three of the world's top AI labs, working separately, all agree: this is permanent.

Every time ChatGPT gives you an answer, ask yourself: is this real, or is it just a confident guess?

[meinenaffenhosen]

For those that like to actually read the paper : https://arxiv.org/pdf/2509.04664

Why Language Models Hallucinate

Spoiler:

[leox-un]

valley of ashes

[killy-deactivated20241228]

[evilwizard]

[parkchanwoohoo]

Incredible

[government-cy-op]

Just found 2 copies of a 19th century naturalism magazine and LOOK AT THEIR LITTLE ICON

[octopushotdogs]

This is the logo of the American Society of Naturalists and they still use it!

[flipchild]

[flipchild]